Analysis of Algorithms CSCI 3110

Previous Evaluations of Programs Correctness – does the algorithm do what it is supposed to do? Generality – does it solve only a specific case or will it work for the general case too? Robustness – will the algorithm recover following an error Portability Documentation & Style Efficiency – Does it accomplish its goals with a minimal usage of computer resources (time and space)?

New Way to Evaluate Memory efficiency: which algorithm requires less memory space during run time? Time efficiency: which algorithm requires less computational time to complete? When comparing the algorithms in terms of efficiency, we consider the amount of memory and time required as a function of the size of the input. Thus we can measure the rate of growth of the memory use or time use using that function.

What is the size of the input? Sorting a set of numbers  the number of items to be sorted Search for an item in a set of items  the number of items to search from Computation involved with a n x n matrix  size of the matrix – A list of strings (numStrings x string Length)

Time Complexity Best time – The minimum amount of time required by the algorithm for any input of size n. – Seldom of interest. Worst time  our focus – The maximum amount of time required by the algorithm for any input of size n Average time – The average amount of time required by the algorithm over all inputs of size n Similar definitions can be given for space complexity

Empirical vs. theoretical analysis Empirical approach: Measure Computer Clock Time – problem  difference in computer platforms, compiler, language, programmers – Machine instructions Theoretical approach: Count number of C++ statements executed during run time or obtain an acceptable approximation. – Therefore, the focus of this type of analysis is performing frequency counts : a count of how many times a statement is executed in an algorithm.

Look at Counting Examples

Algorithm Efficiency The efficiency of an algorithm is determined in terms of the order of growth of the function. Only compare the rate of growth (order of magnitude) of the functions and compare the orders. -- how fast does the algorithm grow as the size of input, N, grows? Normally, the growth function will approach some simpler function asymptotically. For growth functions having different orders of magnitude, exact frequency counts and the constants become insignificant.

The algorithms are grouped into groups according to the order of growth of the their time complexity functions, which include: O(1) < O(logN) < O(N) < O(NlogN) < O(N 2 ) < O(N 3 ) < … < O(e N ) look at the graph with all these functions plotted against each other

Determining the Order of Growth How does one determine which Big O category an algorithm’s growth rate function (GRF) falls into? This is determined by using the big O notation: Rules of Big O notations that help to simplify the analysis of an algorithm – Ignore the low order terms in an algorithm’s growth rate function – Ignore the multiplicative constants in the high order terms of GRF – When combining GRF, O(f(N))+O(g(N)) = O(f(N)+g(N)) (e.g., when two segments of code are analyzed separately and we want to know the time complexity of the total of the two segments

Examples F(n) = 10n + 3n is O(?) F(n) = 20nlogn + 3n + 2 is O(?) F(n) = 12nlogn + 2n 2 is O(?) F(n) = 2 n + 3 n is O(?)

Formal Definition of Big O Formal Definition of Big O function: F(n)= O(g(n)) (F(n) is Big O of g(n)) if there is are positive constant values C and n 0, such that: f(n) = n 0, where n 0 is non-negative integer Note: Some definitions of Big O read that |f(N)| = N 0, where N 0 is non-negative integer. However, we will use the “relaxed” version of the definition since we are dealing with frequency counts which should be positive anyway.

Example Show by definition that f(n) = 3n + 2 is O(n). By definition, we must show there exist C and n 0 such that 3n+2 n 0 where n 0 is a non- negative integer. Let C = 4. Why did I pick 4? If C = 4, 3n+2 < 4n when? Subtract 3n from both sides of the inequality: 3n + 2 – 3n < 4n – 3n or 2 < n. Thus n 0 is 2. Is this the only C and n 0 that will work?

Intuitive interpretation of growth functions A GFR of O(1) implies a problem – whose time requirement is constant & – thus independent of the problem’s size n. A GFR of O(log 2 n) implies a problem – For which the time requirement increases slowly as the problem size increases. – If you square the problem size, you only double its time requirement. – Often occurs in an algorithm that solves a problem by solving a smaller constant fraction of the problem (like the binary search)

Intuitive interpretation continued A GFR of O(1) implies a problem – For which the time requirement is directly proportional to the size of the problem. – In which if you square the problem size, you also square its time requirement – That often has a single loop from 1 to n A GFR of O(n log 2 n) implies a problem – For which the time requirement increases more rapidly than a linear algorithm. – That usually divides a problem into smaller problems that are each solved separately

Intuitive interpretation continued A GFR of O(n 2 ) is a problem – For which the time requirement increases rapidly with the size of the problem – That often has two nested loops A GFR of O(n 3 ) is a problem – That often has three nested loops – that is only practical for small problems A GFR of O(2 n ) is a problem – For which the time requirements are exponential – For which the time requirements increase too rapidly to be practical

Read “Keeping Your Perspective” – pages in the book.